Stripping Alpha Channel from Images

I want to strip the alpha channel (transparent background) from PNGs, and then write them as JPEG images. More correctly, I'd like to make the transparent pixels white. I've tried two techniques, both of which fail in different ways:

Approach 1:

BufferedImage rgbCopy = new BufferedImage(inputImage.getWidth(), inputImage.getHeight(), BufferedImage.TYPE_INT_RGB);
Graphics2D graphics = rgbCopy.createGraphics();
graphics.drawImage(inputImage, 0, 0, Color.WHITE, null);
graphics.dispose();
return rgbCopy;

Result: image has a pink background.

Approach 2:

final WritableRaster raster = inputImage.getRaster();
final WritableRaster newRaster = raster.createWritableChild(0, 0, inputImage.getWidth(), inputImage.getHeight(), 0, 0, new int[]{0, 1, 2});
ColorModel newCM = new ComponentColorModel(inputImage.getColorModel().getColorSpace(), false, false, Transparency.OPAQUE, DataBuffer.TYPE_BYTE);
return new BufferedImage(newCM, newRaster, false, null);

Result: Image has a black background.

In both cases, the input image is a PNG and the output image is written as a JPEG as follows: ImageIO.write(bufferedImage, "jpg", buffer). In case it's relevant: this is on Java 8 and I'm using the twelvemonkeys library to resize the image before writing it as a JPEG.

I've experimented with a number of variations of the above code, with no luck. There are a number of previous questions that suggest the above code, but it doesn't seem to be working in this case.

Solution

Thanks to Rob's answer, we now know why the colors are messed up.

The problem is twofold:

The default JPEGImageWriter that ImageIO uses to write JPEG, does not write JPEGs with alpha in a way other software understands (this is a known issue).
When passing null as the destination to ResampleOp.filter(src, dest) and the filter method is FILTER_TRIANGLE, a new BufferedImage will be created, with alpha (actually, BufferedImage.TYPE_INT_ARGB).

Stripping out the alpha after resampling will work. However, there is another approach that is likely to be faster and save some memory. That is, instead of passing a null destination, pass a BufferedImage of the appropriate size and type:

public static void main(String[] args) throws IOException {
    // Read input
    File input = new File(args[0]);
    BufferedImage inputImage = ImageIO.read(input);

    // Make any transparent parts white
    if (inputImage.getTransparency() == Transparency.TRANSLUCENT) {
        // NOTE: For BITMASK images, the color model is likely IndexColorModel,
        // and this model will contain the "real" color of the transparent parts
        // which is likely a better fit than unconditionally setting it to white.

        // Fill background  with white
        Graphics2D graphics = inputImage.createGraphics();
        try {
            graphics.setComposite(AlphaComposite.DstOver); // Set composite rules to paint "behind"
            graphics.setPaint(Color.WHITE);
            graphics.fillRect(0, 0, inputImage.getWidth(), inputImage.getHeight());
        }
        finally {
            graphics.dispose();
        }
    }

    // Resample to fixed size
    int width = 100;
    int height = 100;

    BufferedImageOp resampler = new ResampleOp(width, height, ResampleOp.FILTER_TRIANGLE);

    // Using explicit destination, resizedImg will be of TYPE_INT_RGB
    BufferedImage resizedImg = resampler.filter(inputImage, new BufferedImage(width, height, BufferedImage.TYPE_INT_RGB));

    // Write output as JPEG
    ImageIO.write(resizedImg, "JPEG", new File(input.getParent(), input.getName().replace('.', '_') + ".jpg"));
}